[CSC 435] Class Tomorrow and Dusty...

[Andrew J. Pounds]

Guys -- we will not be meeting in person tomorrow.  After a COVID 
booster on Friday and a potential exposure this weekend I have to social 
distance (which I really can't do sitting next to you working on code).  
I will, however, be working on a few video lectures that I will deploy 
this week related to shared memory programming (we need to get started 
with that).  That will make up for the missed time in class.

As discussed in class I have moved the due date for the DustyDeck 
assignment from this coming Sunday to the following Friday (Feb 11).  I 
know this is a tough assignment -- but the skills you are learning will 
really pay off later.  As I have said, with this code it is really hard 
to score massive home runs on improvement with the Intel processor.  As 
an example, in my Fortran 90 code I rewrote the inner loop of ID Check 
so it would use a switch -- that improved things a little.  However, in 
another attempt to speed things up I completely rewrote the logic so 
that there was a pre-computed rank 3 tensor controlling the logic in the 
loop without any sort of "if", "switch", or computed goto - just pure 
numerical computation.  With the right hardware, and based on my 
experience, this should be blindingly fast.  However, with Intel 
processors and the GCC compiler this actually slows the code down!  I 
get consistent results -- it just takes twice as long to get them!  That 
will make you cuss...

Just look for modest gains -- anything to make it a little faster.  
Document what you did, make a graph, and move on to something else.


-- 
*/Andrew J. Pounds, Ph.D./*
/Professor of Chemistry and Computer Science/
/Director of the Computational Science Program/
/Mercer University, Macon, GA 31207 (478) 301-5627/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://theochem.mercer.edu/pipermail/csc435/attachments/20220130/c88d474e/attachment.html>